Update inference-cache-config/qwen.json
Browse files- inference-cache-config/qwen.json +104 -0
inference-cache-config/qwen.json
CHANGED
@@ -0,0 +1,104 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"Qwen/Qwen3-8B": [
|
3 |
+
{
|
4 |
+
"batch_size": 1,
|
5 |
+
"sequence_length": 4096,
|
6 |
+
"num_cores": 2,
|
7 |
+
"auto_cast_type": "bf16"
|
8 |
+
},
|
9 |
+
{
|
10 |
+
"batch_size": 4,
|
11 |
+
"sequence_length": 4096,
|
12 |
+
"num_cores": 2,
|
13 |
+
"auto_cast_type": "bf16"
|
14 |
+
},
|
15 |
+
{
|
16 |
+
"batch_size": 8,
|
17 |
+
"sequence_length": 4096,
|
18 |
+
"num_cores": 2,
|
19 |
+
"auto_cast_type": "bf16"
|
20 |
+
},
|
21 |
+
{
|
22 |
+
"batch_size": 1,
|
23 |
+
"sequence_length": 4096,
|
24 |
+
"num_cores": 8,
|
25 |
+
"auto_cast_type": "bf16"
|
26 |
+
},
|
27 |
+
{
|
28 |
+
"batch_size": 32,
|
29 |
+
"sequence_length": 4096,
|
30 |
+
"num_cores": 8,
|
31 |
+
"auto_cast_type": "bf16"
|
32 |
+
}
|
33 |
+
],
|
34 |
+
"Qwen/Qwen3-0.6B": [
|
35 |
+
{
|
36 |
+
"batch_size": 1,
|
37 |
+
"sequence_length": 4096,
|
38 |
+
"num_cores": 2,
|
39 |
+
"auto_cast_type": "bf16"
|
40 |
+
},
|
41 |
+
{
|
42 |
+
"batch_size": 4,
|
43 |
+
"sequence_length": 4096,
|
44 |
+
"num_cores": 2,
|
45 |
+
"auto_cast_type": "bf16"
|
46 |
+
}
|
47 |
+
],
|
48 |
+
"Qwen/Qwen3-1.7B": [
|
49 |
+
{
|
50 |
+
"batch_size": 1,
|
51 |
+
"sequence_length": 4096,
|
52 |
+
"num_cores": 2,
|
53 |
+
"auto_cast_type": "bf16"
|
54 |
+
},
|
55 |
+
{
|
56 |
+
"batch_size": 4,
|
57 |
+
"sequence_length": 4096,
|
58 |
+
"num_cores": 2,
|
59 |
+
"auto_cast_type": "bf16"
|
60 |
+
}
|
61 |
+
],
|
62 |
+
"Qwen/Qwen3-4B": [
|
63 |
+
{
|
64 |
+
"batch_size": 1,
|
65 |
+
"sequence_length": 4096,
|
66 |
+
"num_cores": 2,
|
67 |
+
"auto_cast_type": "bf16"
|
68 |
+
},
|
69 |
+
{
|
70 |
+
"batch_size": 4,
|
71 |
+
"sequence_length": 4096,
|
72 |
+
"num_cores": 2,
|
73 |
+
"auto_cast_type": "bf16"
|
74 |
+
}
|
75 |
+
],
|
76 |
+
"Qwen/Qwen3-14B": [
|
77 |
+
{
|
78 |
+
"batch_size": 1,
|
79 |
+
"sequence_length": 4096,
|
80 |
+
"num_cores": 8,
|
81 |
+
"auto_cast_type": "bf16"
|
82 |
+
},
|
83 |
+
{
|
84 |
+
"batch_size": 16,
|
85 |
+
"sequence_length": 4096,
|
86 |
+
"num_cores": 8,
|
87 |
+
"auto_cast_type": "bf16"
|
88 |
+
}
|
89 |
+
],
|
90 |
+
"Qwen/Qwen3-32B": [
|
91 |
+
{
|
92 |
+
"batch_size": 1,
|
93 |
+
"sequence_length": 4096,
|
94 |
+
"num_cores": 8,
|
95 |
+
"auto_cast_type": "bf16"
|
96 |
+
},
|
97 |
+
{
|
98 |
+
"batch_size": 16,
|
99 |
+
"sequence_length": 4096,
|
100 |
+
"num_cores": 8,
|
101 |
+
"auto_cast_type": "bf16"
|
102 |
+
}
|
103 |
+
]
|
104 |
+
}
|